CN106096326B - A kind of differential evolution Advances in protein structure prediction based on barycenter Mutation Strategy - Google Patents

Abstract

A differential evolution protein structure prediction method based on the centroid mutation strategy. First, the energy values of each conformation are arranged in ascending order, and the average energy error value between each conformation and the lowest energy conformation is calculated; then, some conformations with lower energy are selected to calculate Centroid conformation; finally, judge the search state achieved by the algorithm according to the average energy error value, so as to design different centroid mutation strategies to generate the test conformation, that is, if the average energy error value is greater than the set threshold, design DE/rand-to-centroid The /1 strategy is mutated, and the test conformation is generated by extracting some fragments in the centroid conformation to replace the corresponding fragments in the randomly selected conformation, otherwise, the DE/centroid/2 strategy is designed to mutate, and the centroid conformation is replaced by extracting fragments in the randomly selected conformation The corresponding fragments in generate a test conformation, thereby improving the algorithm search efficiency and prediction accuracy.

Description

A Differential Evolution Protein Structure Prediction Method Based on Centroid Mutation Strategy

technical field

The present invention relates to the fields of biological informatics, intelligent optimization and computer application, in particular to a differential evolution protein structure prediction method based on centroid mutation strategy.

Background technique

In 1953, J.Watson and F.Crick published the DNA molecular double helix structure model in the British "Nature", marking the birth of molecular biology in a true sense; five years later, F.Crick proposed the "Center of Molecular Biology". The idea of "law" reveals the general law of life genetic information transmission. As a key part of this law, the deciphering of the triple genetic code from DNA to protein amino acid sequence (referred to as "the first code") has been completed as early as 1965; however, the folding code from amino acid sequence to spatial structure (referred to as The "Second Code") has not yet been cracked. With the completion of the sequencing of the human genome in 2003, the number of protein amino acid sequences has increased sharply, and the theoretical study of protein folding codes has become a key issue that needs to be solved urgently in the field of protein engineering.

Structural genomics uses experimental means to determine the three-dimensional structure of proteins. X-ray crystallography is by far the most effective method for studying protein structure, and the precision it can achieve is unmatched by any other method. Its main disadvantages are that protein crystals are difficult to cultivate and the period of crystal structure determination is long. The multidimensional NMR method can directly determine the conformation of proteins in solution, but due to the large amount of samples required and the high purity requirements, currently only small molecular proteins can be determined. In general, protein structure experimental determination methods are extremely time-consuming, costly and labor-intensive.

The de novo prediction method is known as the Holy Grail in the field of protein structure prediction. In view of its important biological significance and the complexity of the problem, in 2005, "Science" magazine listed it as one of the 100 most challenging problems to be solved in the current scientific community. one. The following two factors must be considered in protein de novo prediction methods: (1) protein structure energy function; (2) conformational space search method. The first factor is essentially a molecular mechanics problem, mainly to be able to calculate the energy value corresponding to each protein structure. The second factor is essentially a global optimization problem. By choosing an appropriate optimization method, the conformation space is quickly searched to obtain the conformation corresponding to a certain global minimum energy. Among them, protein conformation space optimization belongs to a class of very difficult NP-Hard problems. Population evolution algorithm is an important method to study protein molecular conformation optimization, mainly including differential evolution algorithm (DE), genetic algorithm (GA), particle swarm algorithm (PSO), these algorithms are not only simple in structure, easy to implement, but also strong in robustness , therefore, is often used for global minimum energy conformation search in ab initio prediction methods. However, the population optimization algorithm belongs to a class of stochastic optimization algorithms. The existing literature on protein conformation optimization mainly studies how to jump from one local minimum solution to another local minimum solution, and does not provide a mechanism to effectively use the intelligent information of the population evolution process to guide the search. , resulting in low algorithm efficiency. In addition, affected by selection pressure and genetic drift during random sampling, all individuals in the population will inevitably converge to a certain absorbing state. For optimization problems such as protein conformation, the absorption state is not necessarily the global optimal solution, which affects the prediction accuracy.

Therefore, the existing population-based protein structure prediction methods have shortcomings in search efficiency and prediction accuracy, which need to be improved.

Contents of the invention

In order to overcome the shortcomings of existing protein structure prediction methods in terms of search efficiency and prediction accuracy, the present invention designs a centroid mutation strategy by extracting low-energy conformation information, and at the same time, based on fragment assembly technology, proposes a method with high search efficiency and high prediction accuracy. Gao's method for protein structure prediction based on differential evolution based on centroid mutation strategy.

The technical solution adopted by the present invention to solve its technical problems is:

A differential evolution protein structure prediction method based on centroid mutation strategy, the optimization method includes the following steps:

1) Select the protein force field model, that is, the energy function E(X);

2) given input sequence information;

3) Initialization: Set the population size NP, cross factor CR, maximum number of iterations, and generate the initial conformation population from the input sequence And initialize the number of iterations G=0, where N represents the number of dimensions, represents the N-th dimension element of the i-th conformation ^Ci ;

4) Calculate the energy function value E(C ⁱ ) of each conformation of the current population, i=1, 2, ..., N, and arrange the conformations in ascending order according to the energy value of each conformation in the current population;

5) Find the conformation C _{best with} the lowest energy in the current population, and calculate the average energy error between the energy of other conformations and the energy E(C _best ) of C _best If the number of iterations G=0, let δ _max =δ;

6) For each conformation individual C ⁱ in the population, i∈{1,2,3,…,NP}, let C _target =C ⁱ , C _target represents the target conformation individual, and extract the conformation with lower energy in the current population information, perform the following operations to generate a variant conformation C _mutant :

6.1) Select the top CT conformations Where CT=rand(NP/3,NP/2), rand(NP/3,NP/2) represents a random integer between NP/3 and NP/2, Represents the Nth dimensional element of the mth selected conformation;

6.2) Calculate the centroid conformation C _centroid of the selected CT conformations = (x _centroid,1 ,x _centroid,2 ,...,x _centroid,N ), where the j-th dimension element of the conformation C _centroid j=1,2,...,N;

6.3) Set the sequence length L, randomly generate 4 integers randint1, randint2, randint3 and randint4 between 1 and L, where randint1 and randint2, randint3 and randint4 are different from each other, let a=min(randint1, randint2), b= max(randint1, randint2), c=min(randint3, randint4), d=max(randint3, randint4), where min means taking the minimum value of two numbers, and max means taking the maximum value of two numbers;

6.4) If δ＞0.5δ _max , design DE/rand-to-centroid/strategy for mutation: randomly select two different conformations C ^rand1 and C ^rand2 from the current population, where rand1≠rand2∈[1,NP] , extract the dihedral angle corresponding to the amino acid of the fragment from position a to position b of centroid conformation C _centroid to replace the dihedral angle corresponding to the same position of conformation C ^rand1 , and extract the amino acid of the fragment from position c to position d of conformation C ^rand2 The corresponding dihedral angle replaces the dihedral angle corresponding to the same position of the conformation C ^rand1 , and then the resulting C ^rand1 is fragment assembled to obtain a variant conformation individual C _mutant ;

6.5) If δ≤0.5δ _max , design a DE/centroid/2 strategy for mutation: randomly select two different conformations C ^rand1 and C ^rand2 from the current population, where rand1≠rand2∈[1,NP], extract the conformation The dihedral angle corresponding to the amino acid of the fragment from position a to position b of C ^rand1 replaces the dihedral angle corresponding to the same position of the centroid conformation C _centroid , and uses the dihedral angle corresponding to the amino acid of the fragment from position c to position d on C ^rand2 The face angle is replaced by the dihedral angle corresponding to the same position of the centroid conformation C _centroid , and then the resulting C _centroid is fragment assembled to obtain a variant conformation individual C _mutant ;

7) Perform a crossover operation on the variant conformation C _mutant to generate a test conformation C _trial :

7.1) Randomly generate a decimal number rand3 between 0 and 1;

7.2) If rand3≤CR, then randomly generate an integer rand4 between 1 and L, use the segment rand4 in the variant conformation C _mutant to replace the corresponding segment in the target conformation C _target , thereby generating a test conformation C _trial , if rand3>CR, Then C _trial is directly equal to the variant conformation C _mutant ;

8) Calculate the energy value E(C _trial ) of the test conformation C _trial , if E(C _trial )-E(C _target )<0, it indicates that the test conformation is better than the target conformation, then the test conformation C _trial replaces the target conformation C _target ;

9) Judging whether the termination condition is satisfied, if so, output the result and exit, otherwise return to step 4).

Further, in step 9), after performing steps 6)-8) for each conformation individual in the population, the number of iterations G=G+1, the termination condition is that the number of iterations G reaches the preset value in step 3) The maximum number of iterations.

The technical idea of the present invention is as follows: firstly, arrange in ascending order according to the energy values of each conformation, and calculate the average energy error value between each conformation and the lowest energy conformation; then, select some conformations with lower energy to calculate the centroid conformation; finally, according to the average The energy error value judges the search state achieved by the algorithm, so as to design different centroid mutation strategies to generate the test conformation, that is, if the average energy error value is greater than the set threshold, design the DE/rand-to-centroid/1 strategy to mutate, through Extract some fragments in the centroid conformation to replace the corresponding fragments in the randomly selected conformation to generate a test conformation, otherwise design a DE/centroid/2 strategy to mutate, and generate a test conformation by extracting fragments in the randomly selected conformation to replace the corresponding fragments in the centroid conformation , so as to improve the algorithm search efficiency and prediction accuracy.

The beneficial effects of the present invention are as follows: the centroid conformation is calculated according to the conformation with lower energy, and the centroid mutation strategy is designed to generate the test conformation by extracting the evolution information of the centroid conformation, thereby improving the prediction accuracy; secondly, according to the average energy error value, the judgment algorithm achieves The search state, so as to design a centroid mutation strategy suitable for the corresponding state to generate a test conformation, so as to improve the search efficiency of the algorithm.

Description of drawings

Fig. 1 is a flowchart of the protein structure prediction method in the present invention.

Fig. 2 is a schematic diagram of conformation update of protein 4ICB predicted by the prediction method in the present invention.

Fig. 3 is a conformational distribution diagram obtained when the prediction method of the present invention predicts the protein 4ICB.

Fig. 4 is the three-dimensional structure predicted by the prediction method of the present invention for protein 4ICB.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings.

Referring to Figure 1 and Figure 4, a protein structure prediction method based on centroid mutation strategy differential evolution, including the following steps:

1) Select the protein force field model, that is, the energy function E(X);

2) given input sequence information;

3) Initialization: Set the population size NP, cross factor CR, maximum number of iterations, and generate the initial conformation population from the input sequence And initialize the number of iterations G=0, where N represents the number of dimensions, represents the N-th dimension element of the i-th conformation ^Ci ;

4) Calculate the energy function value E(C ⁱ ) of each conformation of the current population, i=1, 2, ..., N, and arrange the conformations in ascending order according to the energy value of each conformation in the current population;

5) Record the conformation C _{best with} the lowest energy in the current population, and calculate the average energy error between the energy of other conformations and the energy E(C _best ) of C _best If the number of iterations G=0, let δ _max =δ;

6) For each conformation individual C ⁱ in the population, i∈{1,2,3,…,NP}, let C _target =C ⁱ , C _target represents the target conformation individual, and extract the conformation with lower energy in the current population information, perform the following operations to generate a variant conformation C _mutant :

6.1) Select the top CT conformations Where CT=rand(NP/3,NP/2), rand(NP/3,NP/2) represents a random integer between NP/3 and NP/2, Represents the Nth dimensional element of the mth selected conformation;

6.2) Calculate the centroid conformation C _centroid of the selected CT conformations = (x _centroid,1 ,x _centroid,2 ,...,x _centroid,N ), where the j-th dimension element of the conformation C _centroid j=1,2,...,N;

6.3) Set the sequence length L, randomly generate 4 integers randint1, randint2, randint3 and randint4 between 1 and L, where randint1 and randint2, randint3 and randint4 are different from each other, let a=min(randint1, randint2), b= max(randint1, randint2), c=min(randint3, randint4), d=max(randint3, randint4), where min means taking the minimum value of two numbers, and max means taking the maximum value of two numbers;

6.4) If δ＞0.5δ _max , design DE/rand-to-centroid/strategy for mutation: randomly select two different conformations C ^rand1 and C ^rand2 from the current population, where rand1≠rand2∈[1,NP] , extract the dihedral angle corresponding to the amino acid of the fragment from position a to position b of centroid conformation C _centroid to replace the dihedral angle corresponding to the same position of conformation C ^rand1 , and extract the amino acid of the fragment from position c to position d of conformation C ^rand2 The corresponding dihedral angle replaces the dihedral angle corresponding to the same position of the conformation C ^rand1 , and then the resulting C ^rand1 is fragment assembled to obtain a variant conformation individual C _mutant ;

6.5) If δ≤0.5δ _max , design a DE/centroid/2 strategy for mutation: randomly select two different conformations C ^rand1 and C ^rand2 from the current population, where rand1≠rand2∈[1,NP], extract the conformation The dihedral angle corresponding to the amino acid of the fragment from position a to position b of C ^rand1 replaces the dihedral angle corresponding to the same position of the centroid conformation C _centroid , and uses the dihedral angle corresponding to the amino acid of the fragment from position c to position d on C ^rand2 The face angle is replaced by the dihedral angle corresponding to the same position of the centroid conformation C _centroid , and then the resulting C _centroid is fragment assembled to obtain a variant conformation individual C _mutant ;

7) In order to increase the diversity of the population, perform a crossover operation on the variant conformation C _mutant to generate a test conformation C _trial :

7.1) Randomly generate a decimal number rand3 between 0 and 1;

7.2) If rand3≤CR, then randomly generate an integer rand4 between 1 and L, use the segment rand4 in the variant conformation C _mutant to replace the corresponding segment in the target conformation C _target , thereby generating a test conformation C _trial , if rand3>CR, Then C _trial is directly equal to the variant conformation C _mutant ;

8) Calculate the energy value E(C _trial ) of the test conformation C _trial , if E(C _trial )-E(C _target )<0, it indicates that the test conformation is better than the target conformation, then the test conformation C _trial replaces the target conformation C _target ;

9) Judging whether the termination condition is satisfied, if so, output the result and exit, otherwise return to step 4).

In said step 9), after performing steps 6)-8) for each conformation individual in the population, the number of iterations G=G+1, the termination condition is that the number of iterations G reaches the preset maximum iteration in step 3) frequency

In this example, the α-fold protein 4ICB with a sequence length of 76 is an example, a differential evolution protein structure prediction method based on the centroid mutation strategy, which includes the following steps:

1) Select the Rosetta score3 force field model, namely the energy function E(X);

2) input the sequence information of protein 4ICB;

3) Initialization: set population size NP=50, crossover factor CR=0.5, maximum number of iterations is 10000, generate initial conformation population from input sequence And initialize the number of iterations G=0, where N represents the number of dimensions, represents the N-th dimension element of the i-th conformation ^Ci ;

4) Calculate the energy function value E(C ⁱ ) of each conformation of the current population, i=1, 2, ..., N, and arrange the conformations in ascending order according to the energy value of each conformation in the current population;

5) Record the conformation C _{best with} the lowest energy in the current population, and calculate the average energy error between the energy of other conformations and the energy E(C _best ) of C _best If the number of iterations G=0, let δ _max =δ;

6) For each conformation individual C ⁱ in the population, i∈{1,2,3,…,NP}, let C _target =C ⁱ , C _target represents the target conformation individual, and extract the conformation with lower energy in the current population information, perform the following operations to generate a variant conformation C _mutant :

6.1) Select the top CT conformations Where CT=rand(NP/3,NP/2), rand(NP/3,NP/2) represents a random integer between NP/3 and NP/2, Represents the Nth dimensional element of the mth selected conformation;

6.2) Calculate the centroid conformation C _centroid of the selected CT conformations = (x _centroid,1 ,x _centroid,2 ,...,x _centroid,N ), where the j-th dimension element of the conformation C _centroid j=1,2,...,N;

6.3) Set sequence length L=76, randomly generate 4 integers randint1, randint2, randint3 and randint4 between 1 and L, wherein randint1 and randint2, randint3 and randint4 are different from each other, let a=min(randint1, randint2), b=max(randint1, randint2), c=min(randint3, randint4), d=max(randint3, randint4), where min means taking the minimum value of two numbers, and max means taking the maximum value of two numbers;

6.4) If δ＞0.5δ _max , design DE/rand-to-centroid/strategy for mutation: randomly select two different conformations C ^rand1 and C ^rand2 from the current population, where rand1≠rand2∈[1,NP] , extract the dihedral angle corresponding to the amino acid of the fragment from position a to position b of centroid conformation C _centroid to replace the dihedral angle corresponding to the same position of conformation C ^rand1 , and extract the amino acid of the fragment from position c to position d of conformation C ^rand2 The corresponding dihedral angle replaces the dihedral angle corresponding to the same position of the conformation C ^rand1 , and then the resulting C ^rand1 is fragment assembled to obtain a variant conformation individual C _mutant ;

6.5) If δ≤0.5δ _max , design a DE/centroid/2 strategy for mutation: randomly select two different conformations C ^rand1 and C ^rand2 from the current population, where rand1≠rand2∈[1,NP], extract the conformation The dihedral angle corresponding to the amino acid of the fragment from position a to position b of C ^rand1 replaces the dihedral angle corresponding to the same position of the centroid conformation C _centroid , and uses the dihedral angle corresponding to the amino acid of the fragment from position c to position d on C ^rand2 The face angle is replaced by the dihedral angle corresponding to the same position of the centroid conformation C _centroid , and then the resulting C _centroid is fragment assembled to obtain a variant conformation individual C _mutant ;

7) In order to increase the diversity of the population, perform a crossover operation on the variant conformation C _mutant to generate a test conformation C _trial :

7.1) Randomly generate a decimal number rand3 between 0 and 1;

7.2) If rand3≤CR, then randomly generate an integer rand4 between 1 and L, use the segment rand4 in the variant conformation C _mutant to replace the corresponding segment in the target conformation C _target , thereby generating a test conformation C _trial , if rand3>CR, Then C _trial is directly equal to the variant conformation C _mutant ;

8) Calculate the energy value E(C _trial ) of the test conformation C _trial , if E(C _trial )-E(C _target )<0, it indicates that the test conformation is better than the target conformation, then the test conformation C _trial replaces the target conformation C _target ;

9) After executing steps 6)-8) for each conformation individual in the population, the number of iterations G=G+1, if the number of iterations G reaches the maximum number of iterations 10000, output the result and exit, otherwise return to step 4) .

Taking the α-fold protein 4ICB with a sequence length of 76 as an example, the near-native conformation of the protein was obtained by using the above method, and the minimum root mean square deviation is The average root mean square deviation is The predicted three-dimensional structure is shown in Fig. 4.

What has been set forth above is the excellent optimization effect shown by an embodiment of the present invention. Obviously, the present invention is not only suitable for the above-mentioned embodiment, but also can be applied to various fields in actual engineering, while not departing from the basic spirit of the present invention and not exceeding Under the premise of the content involved in the essence of the present invention, various changes can be made to it and implemented.

Claims (2)

1. a kind of differential evolution Advances in protein structure prediction based on barycenter Mutation Strategy, it is characterised in that：The protein Structure Prediction Methods include the following steps：

1) protein force field model, i.e. energy function E (X) are chosen；

2) list entries information is given；

3) it initializes：Population Size NP is set, factor CR is intersected, maximum iteration generates initial configurations kind by list entries GroupAnd initialize iterations G=0, wherein N Representation dimension,Indicate i-th of conformation CⁱN-dimensional element；

4) the energy function value E (C of current each conformation of population are calculatedⁱ), i=1,2 ..., NP, and according to each conformation in current population Energy value carries out ascending order arrangement to each conformation；

5) the conformation C of minimum energy in current population is found out_best, and calculate the energy and C of other conformations_bestENERGY E (C_best) Average energy errorIf iterations G=0, enables δ_max=δ；

6) each conformation individual C being directed in populationⁱ, i ∈ { 1,2,3 ..., NP } enable C_target=Cⁱ, C_targetIndicate target structure As individual, the lower Constellation information of energy in current population is extracted, following operation is executed and generates variation conformation C_mutant：

6.1) CT conformation before the lower ranking of energy is chosen in current population Wherein CT=rand (NP/3, NP/2), rand (NP/3, NP/2) indicate the random integers between NP/3 and NP/2,It indicates The N-dimensional element of m-th of selection conformation；

6.2) the barycenter conformation C of CT selected conformation is calculated_centroid=(x_centroid,1,x_centroid,2,…,x_centroid,N), Wherein, conformation C_centroidJth tie up element

6.3) be arranged sequence length L, generated at random between 1 and L 4 integers randint1, randint2, randint3 and Randint4, wherein randint1 and randint2, randint3 and randint4 are different, enable a=min (randint1, Randint2), b=max (randint1, randint2), c=min (randint3, randint4), d=max (randint3, randint4), wherein min indicate that the minimum value of two numbers, max is taken to indicate to take the maximum value of two numbers；

If 6.4) 0.5 δ of δ ＞_max, then DE/rand-to-centroid/ strategies are designed into row variation：It is random from current population Choose two different conformation C^rand1And C^rand2, wherein rand1 ≠ rand2 ∈ [1, NP], extraction barycenter conformation C_centroidPosition Dihedral angle corresponding to amino acid of a to the segment of position b replaces conformation C^rand1Same position corresponding to dihedral angle, simultaneously Extract conformation C^rand2Dihedral angle corresponding to amino acid of the position c to the segment of position d replaces conformation C^rand1Same position institute is right The dihedral angle answered, then by gained C^rand1Segment is carried out to assemble to obtain variation conformation individual C_mutant；

If 6.5) δ of δ≤0.5_max, then DE/centroid/2 strategies are designed into row variation：Two are randomly selected from current population A different conformation C^rand1And C^rand2, wherein rand1 ≠ rand2 ∈ [1, NP], extraction conformation C^rand1Pieces of the position a to position b Dihedral angle corresponding to the amino acid of section replaces barycenter conformation C_centroidSame position corresponding to dihedral angle, use simultaneously C^rand2Dihedral angle corresponding to the amino acid of segments of the upper position c to position d replaces barycenter conformation C_centroidSame position institute is right The dihedral angle answered, then by gained C_centroidSegment is carried out to assemble to obtain variation conformation individual C_mutant；；

7) to the conformation C that makes a variation_mutantIt executes crossover operation and generates test conformation C_trial：

7.1) decimal rand3 is generated at random between zero and one；

If 7.2) rand3≤CR, integer rand4 is generated at random between 1 and L, utilize variation conformation C_mutantIn segment Rand4 replaces target conformation C_targetIn corresponding segment, to generate test conformation C_trialIf rand3>CR, then C_trialDirectly It connects equal to variation conformation C_mutant；

8) test conformation C is calculated_trialEnergy value E (C_trial), if E (C_trial)-E(C_target) ＜ 0, show that test conformation is excellent In target conformation, then conformation C is tested_trialReplace target conformation C_target；

9) judge whether to meet end condition, result is exported if meeting and exit, otherwise return to step 4).

2. a kind of double-deck differential evolution Advances in protein structure prediction based on barycenter Mutation Strategy as described in claim 1, It is characterized in that：In the step 9), step 6) -8 has been carried out to each conformation individual in population) after, iterations G =G+1, end condition are that iterations G reaches preset maximum iteration in step 3).